This thesis proposes a model of RNA secondary structures with or without pseudoknots. According to a combinatorial approach, we design different models of these structures which we study according to two aspects. In one hand, we define random generation models which allow us to define a measure allowing a better recognition of biological structures. On the other hand, greatings to appropriated encodings and bijections to languages represented by non-contextual grammars, we count the structures composing the space of exact secondary structure prediction algorithms with pseudoknots. The first part deals with random models of RNA structures without pseudoknots. We show that these structures are a relevant source of random noise when determinin...